Active boundary annotation using random MAP perturbations
We address the problem of efficiently annotating labels of objects when they are structured. Often the distribution over labels can be described using a joint potential function over the labels for which sampling is provably hard but efficient maximum a-posteriori (MAP) solvers exist. In this settin...
| Glavni autori: | , , |
|---|---|
| Daljnji autori: | |
| Format: | Članak |
| Jezik: | en_US |
| Izdano: |
PLMR
2018
|
| Online pristup: | http://hdl.handle.net/1721.1/115314 https://orcid.org/0000-0002-2199-0379 |
| _version_ | 1826213782855090176 |
|---|---|
| author | Maji, Subhransu Hazan, Tamir Jaakkola, Tommi S |
| author2 | Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science |
| author_facet | Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Maji, Subhransu Hazan, Tamir Jaakkola, Tommi S |
| author_sort | Maji, Subhransu |
| collection | MIT |
| description | We address the problem of efficiently annotating labels of objects when they are structured. Often the distribution over labels can be described using a joint potential function over the labels for which sampling is provably hard but efficient maximum a-posteriori (MAP) solvers exist. In this setting we develop novel entropy bounds that are based on the expected amount of perturbation
to the potential function that is needed to change MAP decisions. By reasoning about the entropy reduction and cost tradeoff, our algorithm actively selects the next annotation task. As an example of our framework we propose a boundary refinement task which can used to obtain pixelaccurate image boundaries much faster than traditional tools by focussing on parts of the image for refinement in a multi-scale manner. |
| first_indexed | 2024-09-23T15:54:45Z |
| format | Article |
| id | mit-1721.1/115314 |
| institution | Massachusetts Institute of Technology |
| language | en_US |
| last_indexed | 2024-09-23T15:54:45Z |
| publishDate | 2018 |
| publisher | PLMR |
| record_format | dspace |
| spelling | mit-1721.1/1153142022-10-02T05:00:41Z Active boundary annotation using random MAP perturbations Maji, Subhransu Hazan, Tamir Jaakkola, Tommi S Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Jaakkola, Tommi S We address the problem of efficiently annotating labels of objects when they are structured. Often the distribution over labels can be described using a joint potential function over the labels for which sampling is provably hard but efficient maximum a-posteriori (MAP) solvers exist. In this setting we develop novel entropy bounds that are based on the expected amount of perturbation to the potential function that is needed to change MAP decisions. By reasoning about the entropy reduction and cost tradeoff, our algorithm actively selects the next annotation task. As an example of our framework we propose a boundary refinement task which can used to obtain pixelaccurate image boundaries much faster than traditional tools by focussing on parts of the image for refinement in a multi-scale manner. 2018-05-11T14:21:59Z 2018-05-11T14:21:59Z 2014-04 Article http://purl.org/eprint/type/ConferencePaper http://hdl.handle.net/1721.1/115314 Maji, Subhransu, Tamir Hazan, and Tommi Jaakkola. "Active Boundary Annotation using Random MAP Perturbations." Proceedings of the 17th International Conference on Artificial Intelligence and Statistics (AISTATS), 22-25 April, 2014, Reykjavik, Iceland, PMLR, 2014. © The Authors https://orcid.org/0000-0002-2199-0379 en_US http://proceedings.mlr.press/v33/#default Proceedings of the 17th International Conference on Artificial Intelligence and Statistics (AISTATS) Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf PLMR Other univ. web domain |
| spellingShingle | Maji, Subhransu Hazan, Tamir Jaakkola, Tommi S Active boundary annotation using random MAP perturbations |
| title | Active boundary annotation using random MAP perturbations |
| title_full | Active boundary annotation using random MAP perturbations |
| title_fullStr | Active boundary annotation using random MAP perturbations |
| title_full_unstemmed | Active boundary annotation using random MAP perturbations |
| title_short | Active boundary annotation using random MAP perturbations |
| title_sort | active boundary annotation using random map perturbations |
| url | http://hdl.handle.net/1721.1/115314 https://orcid.org/0000-0002-2199-0379 |
| work_keys_str_mv | AT majisubhransu activeboundaryannotationusingrandommapperturbations AT hazantamir activeboundaryannotationusingrandommapperturbations AT jaakkolatommis activeboundaryannotationusingrandommapperturbations |